I'm sure I've seen a statement very similar to that on the forums before. Seems a bit too 'hey isn't ZFS wonderful, you should only use its tools because they're always right'.
I cannot say what was the intention of the author but I had the impression while reading it that his warning was more like "these tools can give wrong answers and should be used with care when dealing with ZFS". Of course, I extracted this quote from a longer chapter and this is may be why this quote gave you this impression.
you could argue the zpool output is 'wrong or confusing'.
That's true, at least for confusion
.
Of course I'd still use the zfs command before df, which was never really designed to handle multiple file systems sharing space.
You are absolutely right. As I just said above, I think this was the message of the author.
For those interested in the exercise, the problem of
df(1) with system sharing space can be reproduced easily. First create three files (three files is just an example, you would see the same behaviour with a mirror or just one data storage).
# mkdir -p /usr/local/fakedisks
# truncate -s 1G /usr/local/fakedisks/disk1
# truncate -s 1G /usr/local/fakedisks/disk2
# truncate -s 1G /usr/local/fakedisks/disk3
Then use them to create a pool.
# zpool create -O canmount=off testpool raidz1 /usr/local/fakedisks/disk1 /usr/local/fakedisks/disk2 /usr/local/fakedisks/disk3
Then create two datasets.
# zfs create -o mountpoint=/mnt/test1 testpool/test1
# zfs create -o mountpoint=/mnt/test2 testpool/test2
Code:
# zfs list testpool
NAME USED AVAIL REFER MOUNTPOINT
testpool 506K 1.93G 24.0K /testpool
The size of the
testpool dataset is equal to
1.9G and to the total space available on the pool.
Now, ask
df(1) the size and available space.
Code:
# df -h mnt/*
Filesystem Size Used Avail Capacity Mounted on
testpool/test1 1.9G 24K 1.9G 0% /mnt/test1
testpool/test2 1.9G 24K 1.9G 0% /mnt/test2
So basicaly, the total size is
3.8G (
1.9G for
test1 plus
1.9G for
test2) which is not quite true. It comes from the fact that when
df(1) ask ZFS the size of each dataset, ZFS
gives the available space on the pool, which is
1.9G, for each dataset.
Now, create a file of
100M in the
test1 folder.
# dd if=/dev/zero of=/mnt/test1/beforesnapshot bs=10M count=10
Check the size and available space again.
Code:
# df -h mnt/*
Filesystem Size Used Avail Capacity Mounted on
testpool/test1 1.9G 100M 1.8G 5% /mnt/test1
testpool/test2 1.8G 24K 1.8G 0% /mnt/test2
Now,
df(1) still shows
1.9G for the size of
test1, but the size of
test2 has shrunk down to
1.8G. If fact, when
df(1) ask ZFS, ZFS
gives the used space (here
100M for
test1 and
0M for
test2) plus the available space on the pool (
1.8G for each) which gives this
1.9G for
test1 versus
1.8G for
test2. So basically, as the datasets fill up, the total size shown by
df(1) shrink.
Another problem arises with snapshots. Let's take a snapshot of
test1 and check with
df(1).
# zfs snapshot testpool/test1@snap1
Code:
# df -h mnt/*
Filesystem Size Used Avail Capacity Mounted on
testpool/test1 1.9G 100M 1.8G 5% /mnt/test1
testpool/test2 1.8G 24K 1.8G 0% /mnt/test2
Nothing changed but this is expected, the dataset does not cost anything in size since there was no modification.
Now, let's imagine that I need space and I decide to remove the
beforesnapshot file.
# rm /mnt/test1/beforesnapshot
Code:
# df -h mnt/*
Filesystem Size Used Avail Capacity Mounted on
testpool/test1 1.8G 24K 1.8G 0% /mnt/test1
testpool/test2 1.8G 24K 1.8G 0% /mnt/test2
While I would expect the available space to go back from
1.8G to
1.9G, instead the
test1 size shrinks from
1.9G to
1.8G. In fact, the space used by the file we just removed is still in use by the snapshot previously taken. So ZFS
tells df(1) that the size of the datatsets
test1 and
test2 is the available space on the pool which is
1.8G in that case.
For at least the behavior described here,
df(1) can be wrong or confusing if one is not careful enough when using it.
Looks interesting, but they only accept paypal. I do not have paypal and look at that also rather skeptical (call me a luddite if you want, I don't like them).
Sorry, no sale
I have a Kobo so I bought it with them using a Credit Card. And I won't call you a luddite (it would be strange coming from someone who does not want to use a smartphone. By the way, you taught me a new word).
